Update app.py

app.py

@@ -1,7 +1,6 @@
-import pandas as pd
-import numpy as np
 import gradio as gr
 import torch
+import numpy as np
 from transformers import AutoModelForMultipleChoice, AutoTokenizer
 
 model_id = "deepset/deberta-v3-large-squad2"
@@ -11,36 +10,51 @@ model = AutoModelForMultipleChoice.from_pretrained(model_id)
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 
 # Define the preprocessing function
-def preprocess(sample):
-    first_sentences = [sample["prompt"]] * 5
-    second_sentences = [sample[option] for option in "ABCDE"]
-    tokenized_sentences = tokenizer(first_sentences, second_sentences, truncation=True, padding=True, return_tensors="pt")
-    sample["input_ids"] = tokenized_sentences["input_ids"]
-    sample["attention_mask"] = tokenized_sentences["attention_mask"]
-    return sample
+def preprocess(text):
+    # Split the input text into lines
+    lines = text.strip().split("\n")
+    samples = []
+
+    # Loop through each line and create a sample
+    for line in lines:
+        parts = line.split("\t")
+        if len(parts) >= 6:
+            sample = {
+                "prompt": parts[0],
+                "A": parts[1],
+                "B": parts[2],
+                "C": parts[3],
+                "D": parts[4],
+                "E": parts[5]
+            }
+            samples.append(sample)
+    
+    return samples
 
 # Define the prediction function
 def predict(data):
-    inputs = torch.stack(data["input_ids"])
-    masks = torch.stack(data["attention_mask"])
-    with torch.no_grad():
-        logits = model(inputs, attention_mask=masks).logits
-    predictions_as_ids = torch.argsort(-logits, dim=1)
-    answers = np.array(list("ABCDE"))[predictions_as_ids.tolist()]
-    return ["".join(i) for i in answers[:, :3]]
-text=gr.Textbox(placeholder="paste multiple choice questions.....")
-label=gr.Label(num_top_classes=3)
+    results = []
+    for sample in data:
+        first_sentences = [sample["prompt"]] * 5
+        second_sentences = [sample[option] for option in "ABCDE"]
+        tokenized_sentences = tokenizer(first_sentences, second_sentences, truncation=True, padding=True, return_tensors="pt")
+        inputs = tokenized_sentences["input_ids"]
+        masks = tokenized_sentences["attention_mask"]
+        with torch.no_grad():
+            logits = model(inputs, attention_mask=masks).logits
+        predictions_as_ids = torch.argsort(-logits, dim=1)
+        answers = np.array(list("ABCDE"))[predictions_as_ids.tolist()]
+        results.append(["".join(i) for i in answers[:, :3]])
+    return results
+
 # Create the Gradio interface
 iface = gr.Interface(
     fn=predict,
-    inputs=text ,  # Use the correct class with type="json"
-    outputs=label,
+    inputs=gr.inputs.Textbox(placeholder="Paste multiple-choice questions (prompt and options separated by tabs, one question per line) ..."),
+    outputs=gr.outputs.Label(num_top_classes=3),
     live=True,
-    examples=[
-        {"prompt": "This is the prompt", "A": "Option A text", "B": "Option B text", "C": "Option C text", "D": "Option D text", "E": "Option E text"}
-    ],
     title="LLM Science Exam Demo",
-    description="Enter the prompt and options (A to E) below and get predictions.",
+    description="Enter multiple-choice questions (prompt and options) below and get predictions.",
 )
 
 # Run the interface